RESEARCH ON EMOTION RECOGNITION METHOD BASED ON BIMODAL SIGNALS OF VISION-EVOKED ELECTROENCEPHALOGRAPHY

RESEARCH ON EMOTION RECOGNITION METHOD BASED ON BIMODAL SIGNALS OF VISION-EVOKED ELECTROENCEPHALOGRAPHY

With the development of the economy, an increasing number of people are choosing cars as their preferred mode of transportation. In this context, the safety of drivers becomes particularly important, as stable emotions can ensure their safety and efficiency during the driving process to the greatest extent possible. This paper explores a driver emotion recognition method based on an improved Triplet Network, addressing the issue of unstable emotions in drivers leading to traffic accidents during driving. The study focuses on the biometric signals of electroencephalography (EEG) and electromyography (EMG), and designs an adaptive bimodal emotion collection experimental paradigm to collect biometric signals under three emotional states: anger, happiness, and normalcy. Subsequently, the ReliefFJMIM feature selection algorithm is utilized to select effective and nonredundant feature sets, followed by feature fusion and normalization. Finally, the LSTM-Triplet emotion classification model is proposed, achieving an accuracy rate of 94.73% in emotion classification and recognition using the de-redundant feature set. Comparative analysis with various existing network model classification methods indicates that the proposed emotion recognition model performs exceptionally well in the field of emotion recognition. It effectively identifies drivers’emotions during the driving process, thereby aiding in the reduction of traffic accidents caused by emotional issues.